Theory And Application Of Multi-scale Multivariate Image Analysis For Machine Vision Inspection

China has become a big country in manufacturing industry. But the industrial inspections in China are still mainly relying on the manual operation of workers which is seriously blocking the development of modern industry. Machine vision inspection is the inevitable trend of industrial automation in the modernization of China. Based on the local and aboard development and application of machine vision, the main contents of this study have focused on the key issues in machine vision inspection, the image measurement and visual inspection, including: planar visual metrology, surface defects detection, color texture surface grading. On the basis of the analysis of perceptual characteristics of observation, physiological structure of the human visual system and mechanism of color texture perception, this study innovatively propose the basic theory of the multi-scale multivariate image analysis for the application of machine vision inspection. The experiments were carried out to verify the proposed methods. The main contents are as follows:In chapter 1, the related study background and significance of the subject are expounded. On the basis of referring to domestic and international associated documents, system components, key technologies and the current application research situation of machine vision inspection are summarized. And the methods of the vision bionics and multivariate image analysis are introduced. Then the main problems and new challenges in the application and research of machine vision inspection are analyzed, and the main work of this study is proposed.In chapter 2, the methods to improve the measurement accuracy are proposed and studied based on the basic the principles of the planar visual metrology. In the linear model of the plane to plane homograph matrix, for the purpose of improving the accuracy of the dynamic image measurement, an image planar measurement method based on the subspace calibration and estimation of homograph matrix is proposed in this study. With the image measurement as the position feedback, the pneumatic control experiments are implemented, the results show that good accuracy and performance can be achieved with the use of the proposed image dynamic measurement method. In the nonlinear distortion model, a partition-based camera intrinsic and extrinsic parameters calibration method for planar visual metrology is proposed. Experiments in measuring the distances of planar objects were carried out to verify the performance of the method.In chapter 3, the principles of the multivariate image analysis (MIA) are studied, and the surface defects detection methods based on MIA with Gaussian multi-scale representations are proposed. With principal component analysis the MIA techniques decompose the Gaussian multi-scale representations into a series of principal components. The Q-statistic image, derived from the residuals after the extraction of the first principal component score and noise, could be used to efficiently reveal the surface defects. This procedure could be regarded as the imitation of human's detection for defects. The comparative experimental results show that the proposed method has more robustness and reliability of defect detection with lower pseudo reject rate.In chapter 4, an integrated quality inspection system for the body of packaging cans is presented for the complicate inspection tasks with the single camera setup including the key components. The machine vision inspection algorithms based on the morphological regions inspection and completeness check, the dynamic local thresholding filter are implemented. The inspection algorithm for inner surface defects detection based on the online multivariate image analysis with the standard defect-free images and the test images are proposed to improve the robustness to illumination. Experimental results show more robustness of the proposed method.In chapter 5, the Gaussian multi-scale multivariate image analysis is extended to color image analysis, and the image algorithms of color MIA with color texture multi-scale representations are proposed. Through the analysis of the human vision system physiological structures and perceptual mechanisms it appears that the proposed color multi-scale MIA could mimic the human visual procedure of trichromatic multi-scale perception and recetive fied opponent color processing. By using the color multi-scale MIA the color texture multi-scale representations are projected onto the eigenspace to obtain the color-texture eigen features consistent with the human visual peception. And a denoising algorithm for color images through the filtering on the eigen space is proposed according to the color multi-scale MIA. Comparative experimential results show the proposed color image denoising method achieves good performance in accord with people's subjective evaluation of image quality.In chapter 6, a surface grading algorithm using the color texture eigen features based on the color multi-scale MIA is proposed for machine vision inspection. The model of the reference eigenspace is built from the characteristic images of the samples. Then the testing images and training images are all projected onto the reference eigenspace to obtain the representative feature clusters. On the eigenspace the Bhattacharyya distance is used to estimate the similarity between the feature clusters of the testing images and the training images followed by a k-NN classifier for surface grading. Classification experiments on the ceramic tiles and bamboo products strips were carried out to verify the validations of the grading approach.In chapter 7, the major work of the study is summarized, and the conclusions and innovations of the study are elaborated. At the same time, future development is predicted in order to provide references for the further research on this project.